Aprendizaje Automático P64

Practica 04

Árboles de decisión, Random Forest (RF) y Máquinas de Soporte Vectorial (SVM)

Fase 2: Entrenamiento de un Random Forest

Autor: Diego Tapia

Importación de librerías

import pandas as pd  
import numpy as np  
import matplotlib.pyplot as plt 

from time import time
from datetime import datetime, timedelta

from sklearn.impute import SimpleImputer
from sklearn.preprocessing import OneHotEncoder, StandardScaler, OrdinalEncoder, MinMaxScaler
from sklearn.compose import make_column_transformer, ColumnTransformer
from sklearn.pipeline import Pipeline
import pickle
import seaborn as sns
import copy

from sklearn.model_selection import train_test_split
from scikeras.wrappers import KerasClassifier, KerasRegressor
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import StratifiedKFold
from sklearn.model_selection import ShuffleSplit
from sklearn.model_selection import GridSearchCV
from keras.layers import Dropout

import keras
from keras.models import model_from_json
from keras.models import load_model

from sklearn import metrics 
from sklearn.metrics import confusion_matrix  
from sklearn.metrics import accuracy_score
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score

from sklearn import set_config   
from sklearn.ensemble import RandomForestClassifier

print("Módulos y clases importados")

2024-06-14 14:40:15.806095: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-06-14 14:40:17.628217: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

Módulos y clases importados

Carga del dataset

Se carga el dataset procesado de la fase 1

dataframe = pd.read_csv('dataset/dataset_processed_heart_failure.csv', sep=',', header=0)

dataframe.head()

	age	anaemia	creatinine_phosphokinase	diabetes	ejection_fraction	high_blood_pressure	platelets	serum_creatinine	serum_sodium	sex	smoking	time	Y
0	55.0	0	748	0	45	0	263358.03	1.3	137	male	1	88	0
1	65.0	0	56	0	25	0	305000.00	5.0	130	male	0	207	0
2	45.0	0	582	1	38	0	319000.00	0.9	140	female	0	244	0
3	60.0	1	754	1	40	1	328000.00	1.2	126	male	0	90	0
4	95.0	1	582	0	30	0	461000.00	2.0	132	male	0	50	1

Se extrae la variable de salida 'Y'

Y = dataframe['Y']
dataframe.drop(['Y'], axis=1, inplace=True)
dataframe.head()

	age	anaemia	creatinine_phosphokinase	diabetes	ejection_fraction	high_blood_pressure	platelets	serum_creatinine	serum_sodium	sex	smoking	time
0	55.0	0	748	0	45	0	263358.03	1.3	137	male	1	88
1	65.0	0	56	0	25	0	305000.00	5.0	130	male	0	207
2	45.0	0	582	1	38	0	319000.00	0.9	140	female	0	244
3	60.0	1	754	1	40	1	328000.00	1.2	126	male	0	90
4	95.0	1	582	0	30	0	461000.00	2.0	132	male	0	50

Carga del Pipeline

Se carga el pipeline obtenido en la fase 1

def cargarPipeline(nombreArchivo):
    with open(nombreArchivo+'.pickle', 'rb') as handle:
        pipeline = pickle.load(handle)
    return pipeline

nombreArchivoPreprocesador='pipeline_heart_failure'
pipe=None
pipe=cargarPipeline(nombreArchivoPreprocesador) #Debes incluir el archivo del Pipeline en el mismo directorio del script 
cantidadPasos=len(pipe.steps)
print("Cantidad de pasos: ",cantidadPasos)
set_config(display='diagram')
pipe

Cantidad de pasos:  2

/home/diego/UPS/MachineLearning/Unidad 2/Practica04_RandonForest_SVM/mlvenv/lib/python3.12/site-packages/sklearn/compose/_column_transformer.py:1624: FutureWarning: 
The format of the columns of the 'remainder' transformer in ColumnTransformer.transformers_ will change in version 1.7 to match the format of the other transformers.
At the moment the remainder columns are stored as indices (of type int). With the same ColumnTransformer configuration, in the future they will be stored as column names (of type str).
To use the new behavior now and suppress this warning, use ColumnTransformer(force_int_remainder_cols=False).

  warnings.warn(

Pipeline(steps=[('prepcn',
                 ColumnTransformer(n_jobs=2, remainder='passthrough',
                                   transformers=[('catnom',
                                                  Pipeline(steps=[('onehot',
                                                                   OneHotEncoder(handle_unknown='ignore',
                                                                                 sparse_output=False))]),
                                                  ['sex'])])),
                ('prepminmax',
                 ColumnTransformer(remainder='passthrough',
                                   transformers=[('tranminmax',
                                                  Pipeline(steps=[('minmax',
                                                                   MinMaxScaler())]),
                                                  [0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
                                                   10, 11, 12])]))])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Creación del Random Forest

random_forest = RandomForestClassifier(n_estimators=200, max_depth=25, random_state=42)

pipe.steps.append(['modelRandomForest',random_forest])

print("Cantidad de pasos: ",len(pipe.steps))

Cantidad de pasos:  3

Entrenamiento del modelo

# Se divide el dataset en train y test
X = copy.deepcopy(dataframe)

X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size= 0.2, random_state=42)

print("Conjuntos train y test creados")

Conjuntos train y test creados

modelo_entrenado = pipe.fit(X_train, y_train)

print("Entrenamiento terminado")

Entrenamiento terminado

Predicción con modelo base y evaluación de calidad: MAE, MSE, RMSE, Accuracy, Precision, Recall y F1

y_pred = pipe.predict(X_test)
#y_pred

dataframeFinal=pd.DataFrame({'real':y_test, 'predicción': y_pred})

np.set_printoptions(formatter={'float': lambda X: "{0:0.0f}".format(X)})

dataframeFinal.head(20)

	real	predicción
1501	0	0
2586	1	1
2653	1	1
1055	1	1
705	0	0
106	0	0
589	0	0
2468	0	0
2413	0	0
1600	0	0
2464	0	0
228	0	0
915	1	1
794	1	1
3021	0	0
3543	1	1
1073	1	1
3351	0	0
1744	0	0
1084	0	0

from sklearn import metrics 
from sklearn.metrics import confusion_matrix  
from sklearn.metrics import accuracy_score

MAE=metrics.mean_absolute_error(y_test, y_pred)
MSE=metrics.mean_squared_error(y_test, y_pred)
RMSE=np.sqrt(metrics.mean_squared_error(y_test, y_pred))
accuracy=str(accuracy_score(y_test, y_pred))

MAE=str(round(MAE, 4))
MSE=str(round(MSE, 4))
RMSE=str(round(RMSE, 4))

print('Mean Absolute Error (MAE):', MAE)  
print('Mean Squared Error (MSE):', MSE)  
print('Root Mean Squared Error (RMSE):', RMSE) 
print('Accuracy: ' + accuracy) 

print('Confusion_matrix:')
y_test_transformado = y_test
y_pred_transformado = y_pred

cm = confusion_matrix(y_test_transformado, y_pred_transformado)  
print(cm)  
tn, fp, fn, tp = confusion_matrix(y_test_transformado, y_pred_transformado).ravel()  
print(tn,fp,fn,tp)

precision=precision_score(y_test_transformado, y_pred_transformado)
precision=str(round(precision, 4))
print('Precision:',precision)

recall=recall_score(y_test_transformado, y_pred_transformado)
recall=str(round(recall, 4))
print('Recall:',recall)

f1=f1_score(y_test_transformado, y_pred_transformado)
f1=str(round(f1, 4))
print('F1:',f1)

Mean Absolute Error (MAE): 0.007
Mean Squared Error (MSE): 0.007
Root Mean Squared Error (RMSE): 0.0837
Accuracy: 0.993
Confusion_matrix:
[[693   5]
 [  2 300]]
693 5 2 300
Precision: 0.9836
Recall: 0.9934
F1: 0.9885

Tabla 1. Métricas de evaluación Random Forest

Métrica	Valor
Accuracy	0.993
Precision	0.9836
Recall	0.9934
F1	0.9885

Prediccion

def predecirNuevoPaciente(age='86', anaemia=1, creatinine_phosphokinase=682, diabetes=1,
                            ejection_fraction=45, high_blood_pressure=1, platelets=310000, serum_creatinine=4, 
                            serum_sodium=135, sex='female', smoking=1, time=90):   
     
    cnames=['age', 'anaemia', 'creatinine_phosphokinase', 'diabetes', 'ejection_fraction', 'high_blood_pressure', 'platelets', 'serum_creatinine', 'serum_sodium', 'sex' ,'smoking', 'time']
    Xnew=[age, anaemia, creatinine_phosphokinase, diabetes, ejection_fraction, high_blood_pressure, platelets, serum_creatinine, serum_sodium, sex, smoking, time]
    Xnew_Dataframe = pd.DataFrame(data=[Xnew],columns=cnames)
    
    
    pred = (pipe.predict(Xnew_Dataframe) > 0.5).astype("int32")
    pred = pred.flatten()[0]# de 2D a 1D
    return pred

prediccion = predecirNuevoPaciente()
print(f"Prediccion : {prediccion}")
if(prediccion == 0):
    print("El paciente sobrevive")
else:
    print("El paciente fallece")

Prediccion : 1
El paciente fallece

Visualizacion de los arboles de decisión

pipe

/home/diego/UPS/MachineLearning/Unidad 2/Practica04_RandonForest_SVM/mlvenv/lib/python3.12/site-packages/sklearn/compose/_column_transformer.py:1624: FutureWarning: 
The format of the columns of the 'remainder' transformer in ColumnTransformer.transformers_ will change in version 1.7 to match the format of the other transformers.
At the moment the remainder columns are stored as indices (of type int). With the same ColumnTransformer configuration, in the future they will be stored as column names (of type str).
To use the new behavior now and suppress this warning, use ColumnTransformer(force_int_remainder_cols=False).

  warnings.warn(

Pipeline(steps=[('prepcn',
                 ColumnTransformer(n_jobs=2, remainder='passthrough',
                                   transformers=[('catnom',
                                                  Pipeline(steps=[('onehot',
                                                                   OneHotEncoder(handle_unknown='ignore',
                                                                                 sparse_output=False))]),
                                                  ['sex'])])),
                ('prepminmax',
                 ColumnTransformer(remainder='passthrough',
                                   transformers=[('tranminmax',
                                                  Pipeline(steps=[('minmax',
                                                                   MinMaxScaler())]),
                                                  [0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
                                                   10, 11, 12])])),
                ['modelRandomForest',
                 RandomForestClassifier(max_depth=25, n_estimators=200,
                                        random_state=42)]])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Figura 1. Arbol de decisión #10 del Random Forest

from sklearn.tree import export_graphviz
import pydotplus
from IPython.display import Image
import graphviz

forest = pipe.named_steps['modelRandomForest']
estimator = forest.estimators_[10]
#sex_female,sex_male,age,anaemia,creatinine_phosphokinase,diabetes,ejection_fraction,high_blood_pressure,platelets,serum_creatinine,serum_sodium,smoking,time
cnames=['sex_female','sex_male','age', 'anaemia', 'creatinine_phosphokinase', 'diabetes', 'ejection_fraction', 'high_blood_pressure', 'platelets', 'serum_creatinine', 'serum_sodium','smoking', 'time']
dot_data = export_graphviz(estimator, 
                           out_file=None, 
                           feature_names=cnames,  
                           class_names=['Sobrevive', 'Fallece'],  
                           filled=True, 
                           rounded=True,  
                           special_characters=True)  

# Visualizar el árbol
graph = pydotplus.graph_from_dot_data(dot_data)  
Image(graph.create_png())

Tabla 2. Caracteristicas mas importantes del Random Forest

# Obtener la importancia de las características
importances = forest.feature_importances_

# Crear un DataFrame para visualizar la importancia de las características
feature_names = cnames
feature_importances = pd.DataFrame(importances, index=feature_names, columns=['importance']).sort_values('importance', ascending=False)

feature_importances.head()

	importance
time	0.378367
serum_creatinine	0.144224
ejection_fraction	0.118229
age	0.088271
creatinine_phosphokinase	0.071915

Los árboles de decisión del random forest fueron definidos con una profundidad máxima de 25, como se puede observar en la Figura 1. Por efectos prácticos se va a entrenar nuevamente un Random Forest pero ahora con una profundidad máxima de 6 para una mejor visualización y explicación.

Entrenamiento de random forest con profundidad de 6

random_forest_reducido = RandomForestClassifier(n_estimators=20, max_depth=6, random_state=42)

pipe.steps.pop()
print("Cantidad de pasos: ",len(pipe.steps))
pipe.steps.append(['modelRandomForestReduced',random_forest_reducido])
print("Cantidad de pasos: ",len(pipe.steps))
pipe

Cantidad de pasos:  2
Cantidad de pasos:  3

Pipeline(steps=[('prepcn',
                 ColumnTransformer(n_jobs=2, remainder='passthrough',
                                   transformers=[('catnom',
                                                  Pipeline(steps=[('onehot',
                                                                   OneHotEncoder(handle_unknown='ignore',
                                                                                 sparse_output=False))]),
                                                  ['sex'])])),
                ('prepminmax',
                 ColumnTransformer(remainder='passthrough',
                                   transformers=[('tranminmax',
                                                  Pipeline(steps=[('minmax',
                                                                   MinMaxScaler())]),
                                                  [0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
                                                   10, 11, 12])])),
                ['modelRandomForestReduced',
                 RandomForestClassifier(max_depth=6, n_estimators=20,
                                        random_state=42)]])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Entrenamiento modelo Random Forest Reducido

modelo_reducido = pipe.fit(X_train, y_train)

print("Entrenamiento terminado")

Entrenamiento terminado

Evaluación modelo reducido

y_pred = pipe.predict(X_test)

dataframeFinal=pd.DataFrame({'real':y_test, 'predicción': y_pred})

np.set_printoptions(formatter={'float': lambda X: "{0:0.0f}".format(X)})

dataframeFinal.head(20)

	real	predicción
1501	0	0
2586	1	1
2653	1	0
1055	1	1
705	0	0
106	0	0
589	0	0
2468	0	0
2413	0	0
1600	0	0
2464	0	0
228	0	0
915	1	1
794	1	1
3021	0	0
3543	1	1
1073	1	1
3351	0	0
1744	0	0
1084	0	0

from sklearn import metrics 
from sklearn.metrics import confusion_matrix  
from sklearn.metrics import accuracy_score

MAE=metrics.mean_absolute_error(y_test, y_pred)
MSE=metrics.mean_squared_error(y_test, y_pred)
RMSE=np.sqrt(metrics.mean_squared_error(y_test, y_pred))
accuracy=str(accuracy_score(y_test, y_pred))

MAE=str(round(MAE, 4))
MSE=str(round(MSE, 4))
RMSE=str(round(RMSE, 4))

print('Mean Absolute Error (MAE):', MAE)  
print('Mean Squared Error (MSE):', MSE)  
print('Root Mean Squared Error (RMSE):', RMSE) 
print('Accuracy: ' + accuracy) 

print('Confusion_matrix:')
y_test_transformado = y_test
y_pred_transformado = y_pred

cm = confusion_matrix(y_test_transformado, y_pred_transformado)  
print(cm)  
tn, fp, fn, tp = confusion_matrix(y_test_transformado, y_pred_transformado).ravel()  
print(tn,fp,fn,tp)

precision=precision_score(y_test_transformado, y_pred_transformado)
precision=str(round(precision, 4))
print('Precision:',precision)

recall=recall_score(y_test_transformado, y_pred_transformado)
recall=str(round(recall, 4))
print('Recall:',recall)

f1=f1_score(y_test_transformado, y_pred_transformado)
f1=str(round(f1, 4))
print('F1:',f1)

Mean Absolute Error (MAE): 0.055
Mean Squared Error (MSE): 0.055
Root Mean Squared Error (RMSE): 0.2345
Accuracy: 0.945
Confusion_matrix:
[[687  11]
 [ 44 258]]
687 11 44 258
Precision: 0.9591
Recall: 0.8543
F1: 0.9037

Tabla 3. Métricas de Evaluación Random Forest reducido

Métrica	Valor
Accuracy	0.945
Precision	0.9591
Recall	0.8543
F1	0.9037

Si bien los nuevos valores de las métricas de evaluación con un nuevo random forest “reducido” son inferiores al primero, se tiene valores bastante altos.

Visualización del random forest reducido

Figura 2. Arbole de decisión #1 del random forest reducido

from sklearn.tree import export_graphviz
import pydotplus
from IPython.display import Image
import graphviz

forest_reduced = pipe.named_steps['modelRandomForestReduced']
estimator = forest_reduced.estimators_[1]
#sex_female,sex_male,age,anaemia,creatinine_phosphokinase,diabetes,ejection_fraction,high_blood_pressure,platelets,serum_creatinine,serum_sodium,smoking,time
cnames=['sex_female','sex_male','age', 'anaemia', 'creatinine_phosphokinase', 'diabetes', 'ejection_fraction', 'high_blood_pressure', 'platelets', 'serum_creatinine', 'serum_sodium','smoking', 'time']
dot_data = export_graphviz(estimator, 
                           out_file=None, 
                           feature_names=cnames,  
                           class_names=['Fallece', 'Sobrevive'],  
                           filled=True, 
                           rounded=True,  
                           special_characters=True)  

# Visualizar el árbol
graph = pydotplus.graph_from_dot_data(dot_data)  
Image(graph.create_png())

Tabla 4 Características más importantes del random forest (profundidad máxima: 6)

# Obtener la importancia de las características
importances = forest.feature_importances_

# Crear un DataFrame para visualizar la importancia de las características
feature_names = cnames
feature_importances = pd.DataFrame(importances, index=feature_names, columns=['importance']).sort_values('importance', ascending=False)

feature_importances.head()

	importance
time	0.378367
serum_creatinine	0.144224
ejection_fraction	0.118229
age	0.088271
creatinine_phosphokinase	0.071915

Conclusiones

Las características más importantes son las mismas para los dos Random forest entrenados, además como se ve en las Tablas 1 y 3, las metricas de evaluacion no difieren demasiado, por lo que se puede concluir que no hace falta usar grandes cantidades de estimadores y niveles de profundidad al menos para este dataset.

Para encontrar los mejores valores se recomienda un fine tunning.

Con niveles de profundidad bajos, es más fácil explicar los árboles de decisión.

Referencias

Heart failure prediction - clinical records . (2024, May 5). Kaggle. https://www.kaggle.com/datasets/aadarshvelu/heart-failure-prediction-clinical-records

Ortiz, I. R. H., PhD. (2024, January 16). Clasificación y optimización (fine tuning) con RANDOM FOREST para aprobación de un crédito en una institución financiera. Transformación de variables categóricas - GenSciNet. GenSciNet. https://genscinet.com/clasificacion-random-forest-prestamos/

Heart Failure Clinical Records. (2020). UCI Machine Learning Repository. https://doi.org/10.24432/C5Z89R.